Skip to content

Instantly share code, notes, and snippets.

View MichalMalyska's full-sized avatar

Michal Malyska MichalMalyska

35 followers · 86 following
View GitHub Profile
@MichalMalyska
MichalMalyska / config_clean_medium.jsonnet
Created April 29, 2020 18:21
local experiment_name = "cnn_edss19";
{ "train_options": {
"serialization_dir": "/results/dev/" + experiment_name,
"file_friendly_logging": false,
"recover": false,
"force": true,
"node_rank": 0,
"batch_weight_key": "",
"dry_run": false,
@MichalMalyska
MichalMalyska / dataset_readers.py
Created April 28, 2020 21:04
import os
import pandas as pd
from typing import Dict, List, Iterator, Tuple, Union
import logging
import torch
from overrides import overrides
from transformers import BertTokenizerFast
@MichalMalyska
MichalMalyska / MS-BERT Dataset Reader
Last active April 29, 2020 20:10
"dataset_reader": {
"type": "data_scripts.dataset_reader.ms_edss19_reader",
"token_indexers": {
"tokens": {
"type": "pretrained_transformer",
"model_name": "ms-bert",
"namespace": "tokens",
"max_length": 768,
}
},
@MichalMalyska
MichalMalyska / How to use MS-BERT
Created April 28, 2020 20:57
import transformers
tokenizer = AutoTokenizer.from_pretrained("NLP4H/ms_bert")
model = AutoModel.from_pretrained("NLP4H/ms_bert")
@MichalMalyska
MichalMalyska / How to use MS-BERT
Created April 28, 2020 20:56
import transformers
tokenizer = AutoTokenizer.from_pretrained("NLP4H/ms_bert")
model = AutoModelWithLMHead.from_pretrained("NLP4H/ms_bert")
@MichalMalyska
MichalMalyska / CNN Encoder Config MS-BERT
Created April 28, 2020 20:32
"seq2vec_encoder": {
"type": "cnn",
"embedding_dim": hidden_size,
"num_filters": 128,
"ngram_filter_sizes": [2, 3, 4, 5, 6, 10],
@MichalMalyska
MichalMalyska / keybase.md
Created October 17, 2016 18:14

Keybase proof

I hereby claim:

  • I am MichalMalyska on github.
  • I am michalmalyska (https://keybase.io/michalmalyska) on keybase.
  • I have a public key whose fingerprint is 5260 6F1A B94A 5821 2AB9 B502 DAFB 65D5 DA70 D72A

To claim this, I am signing this object: